>>2.1) However, Bates et al have mentioned the restricted likelihood
>>ratio test. There is a package in R implementing some of these tools
>>(RLRsim), but these appear to be limited to and or focused on tests of
>>random effects.
>>
> BMB: You can test fixed effects, apparently, but only *in combination
with*
>a test of the random effect (the null hypothesis is always a model
>without random effects). Also limited to LMMs, and a single
>random effect.
Let me clarify this, since I'm the person who wrote RLRSim and apparently
failed miserably to write useful documentation:
RLRSim implements restricted likelihood ratio tests in LMMs with i.i.d.
errors for testing whether the variance of a single random effect is zero.
This is not limited to models with a single random effect,
BUT the effect that is tested has to be uncorrelated to all the other
random effects in the model.
BMB is absolutely correct as far as likelihood ratio tests for fixed effects
in combination with a random effect are concerned, although I'd want to
caution potential users that those are not as well tested.
Good luck with your course,
Fabian
[[alternative HTML version deleted]]